2 * linux/fs/proc/array.c
4 * Copyright (C) 1992 by Linus Torvalds
5 * based on ideas by Darren Senn
8 * Michael. K. Johnson: stat,statm extensions.
9 * <johnsonm@stolaf.edu>
11 * Pauline Middelink : Made cmdline,envline only break at '\0's, to
12 * make sure SET_PROCTITLE works. Also removed
13 * bad '!' which forced address recalculation for
14 * EVERY character on the current page.
15 * <middelin@polyware.iaf.nl>
17 * Danny ter Haar : added cpuinfo
20 * Alessandro Rubini : profile extension.
21 * <rubini@ipvvis.unipv.it>
23 * Jeff Tranter : added BogoMips field to cpuinfo
24 * <Jeff_Tranter@Mitel.COM>
26 * Bruno Haible : remove 4K limit for the maps file
27 * <haible@ma2s2.mathematik.uni-karlsruhe.de>
29 * Yves Arrouye : remove removal of trailing spaces in get_array.
30 * <Yves.Arrouye@marin.fdn.fr>
32 * Jerome Forissier : added per-CPU time information to /proc/stat
33 * and /proc/<pid>/cpu extension
34 * <forissier@isia.cma.fr>
35 * - Incorporation and non-SMP safe operation
36 * of forissier patch in 2.1.78 by
37 * Hans Marcus <crowbar@concepts.nl>
39 * aeb@cwi.nl : /proc/partitions
42 * Alan Cox : security fixes.
43 * <alan@lxorguk.ukuu.org.uk>
45 * Al Viro : safe handling of mm_struct
47 * Gerhard Wichert : added BIGMEM support
48 * Siemens AG <Gerhard.Wichert@pdb.siemens.de>
50 * Al Viro & Jeff Garzik : moved most of the thing into base.c and
51 * : proc_misc.c. The rest may eventually go into
55 #include <linux/types.h>
56 #include <linux/errno.h>
57 #include <linux/time.h>
58 #include <linux/kernel.h>
59 #include <linux/kernel_stat.h>
60 #include <linux/tty.h>
61 #include <linux/string.h>
62 #include <linux/mman.h>
63 #include <linux/proc_fs.h>
64 #include <linux/ioport.h>
65 #include <linux/uaccess.h>
68 #include <linux/hugetlb.h>
69 #include <linux/pagemap.h>
70 #include <linux/swap.h>
71 #include <linux/smp.h>
72 #include <linux/signal.h>
73 #include <linux/highmem.h>
74 #include <linux/file.h>
75 #include <linux/fdtable.h>
76 #include <linux/times.h>
77 #include <linux/cpuset.h>
78 #include <linux/rcupdate.h>
79 #include <linux/delayacct.h>
80 #include <linux/seq_file.h>
81 #include <linux/pid_namespace.h>
82 #include <linux/ptrace.h>
83 #include <linux/tracehook.h>
84 #include <linux/string_helpers.h>
85 #include <linux/user_namespace.h>
87 #include <asm/pgtable.h>
88 #include <asm/processor.h>
91 static inline void task_name(struct seq_file
*m
, struct task_struct
*p
)
95 char tcomm
[sizeof(p
->comm
)];
98 get_task_comm(tcomm
, p
);
100 seq_puts(m
, "Name:\t");
102 size
= seq_get_buf(m
, &buf
);
103 ret
= string_escape_str(tcomm
, buf
, size
, ESCAPE_SPACE
| ESCAPE_SPECIAL
, "\n\\");
104 seq_commit(m
, ret
< size
? ret
: -1);
110 * The task state array is a strange "bitmap" of
111 * reasons to sleep. Thus "running" is zero, and
112 * you can test for combinations of others with
115 static const char * const task_state_array
[] = {
116 "R (running)", /* 0 */
117 "S (sleeping)", /* 1 */
118 "D (disk sleep)", /* 2 */
119 "T (stopped)", /* 4 */
120 "t (tracing stop)", /* 8 */
122 "Z (zombie)", /* 32 */
125 static inline const char *get_task_state(struct task_struct
*tsk
)
127 unsigned int state
= (tsk
->state
| tsk
->exit_state
) & TASK_REPORT
;
130 * Parked tasks do not run; they sit in __kthread_parkme().
131 * Without this check, we would report them as running, which is
132 * clearly wrong, so we report them as sleeping instead.
134 if (tsk
->state
== TASK_PARKED
)
135 state
= TASK_INTERRUPTIBLE
;
137 BUILD_BUG_ON(1 + ilog2(TASK_REPORT
) != ARRAY_SIZE(task_state_array
)-1);
139 return task_state_array
[fls(state
)];
142 static inline void task_state(struct seq_file
*m
, struct pid_namespace
*ns
,
143 struct pid
*pid
, struct task_struct
*p
)
145 struct user_namespace
*user_ns
= seq_user_ns(m
);
146 struct group_info
*group_info
;
148 struct task_struct
*tracer
;
149 const struct cred
*cred
;
150 pid_t ppid
, tpid
= 0, tgid
, ngid
;
151 unsigned int max_fds
= 0;
154 ppid
= pid_alive(p
) ?
155 task_tgid_nr_ns(rcu_dereference(p
->real_parent
), ns
) : 0;
157 tracer
= ptrace_parent(p
);
159 tpid
= task_pid_nr_ns(tracer
, ns
);
161 tgid
= task_tgid_nr_ns(p
, ns
);
162 ngid
= task_numa_group_id(p
);
163 cred
= get_task_cred(p
);
167 max_fds
= files_fdtable(p
->files
)->max_fds
;
178 "Uid:\t%d\t%d\t%d\t%d\n"
179 "Gid:\t%d\t%d\t%d\t%d\n"
180 "FDSize:\t%d\nGroups:\t",
182 tgid
, ngid
, pid_nr_ns(pid
, ns
), ppid
, tpid
,
183 from_kuid_munged(user_ns
, cred
->uid
),
184 from_kuid_munged(user_ns
, cred
->euid
),
185 from_kuid_munged(user_ns
, cred
->suid
),
186 from_kuid_munged(user_ns
, cred
->fsuid
),
187 from_kgid_munged(user_ns
, cred
->gid
),
188 from_kgid_munged(user_ns
, cred
->egid
),
189 from_kgid_munged(user_ns
, cred
->sgid
),
190 from_kgid_munged(user_ns
, cred
->fsgid
),
193 group_info
= cred
->group_info
;
194 for (g
= 0; g
< group_info
->ngroups
; g
++)
196 from_kgid_munged(user_ns
, GROUP_AT(group_info
, g
)));
200 seq_puts(m
, "\nNStgid:");
201 for (g
= ns
->level
; g
<= pid
->level
; g
++)
202 seq_printf(m
, "\t%d",
203 task_tgid_nr_ns(p
, pid
->numbers
[g
].ns
));
204 seq_puts(m
, "\nNSpid:");
205 for (g
= ns
->level
; g
<= pid
->level
; g
++)
206 seq_printf(m
, "\t%d",
207 task_pid_nr_ns(p
, pid
->numbers
[g
].ns
));
208 seq_puts(m
, "\nNSpgid:");
209 for (g
= ns
->level
; g
<= pid
->level
; g
++)
210 seq_printf(m
, "\t%d",
211 task_pgrp_nr_ns(p
, pid
->numbers
[g
].ns
));
212 seq_puts(m
, "\nNSsid:");
213 for (g
= ns
->level
; g
<= pid
->level
; g
++)
214 seq_printf(m
, "\t%d",
215 task_session_nr_ns(p
, pid
->numbers
[g
].ns
));
220 void render_sigset_t(struct seq_file
*m
, const char *header
,
232 if (sigismember(set
, i
+1)) x
|= 1;
233 if (sigismember(set
, i
+2)) x
|= 2;
234 if (sigismember(set
, i
+3)) x
|= 4;
235 if (sigismember(set
, i
+4)) x
|= 8;
236 seq_printf(m
, "%x", x
);
242 static void collect_sigign_sigcatch(struct task_struct
*p
, sigset_t
*ign
,
245 struct k_sigaction
*k
;
248 k
= p
->sighand
->action
;
249 for (i
= 1; i
<= _NSIG
; ++i
, ++k
) {
250 if (k
->sa
.sa_handler
== SIG_IGN
)
252 else if (k
->sa
.sa_handler
!= SIG_DFL
)
257 static inline void task_sig(struct seq_file
*m
, struct task_struct
*p
)
260 sigset_t pending
, shpending
, blocked
, ignored
, caught
;
262 unsigned long qsize
= 0;
263 unsigned long qlim
= 0;
265 sigemptyset(&pending
);
266 sigemptyset(&shpending
);
267 sigemptyset(&blocked
);
268 sigemptyset(&ignored
);
269 sigemptyset(&caught
);
271 if (lock_task_sighand(p
, &flags
)) {
272 pending
= p
->pending
.signal
;
273 shpending
= p
->signal
->shared_pending
.signal
;
274 blocked
= p
->blocked
;
275 collect_sigign_sigcatch(p
, &ignored
, &caught
);
276 num_threads
= get_nr_threads(p
);
277 rcu_read_lock(); /* FIXME: is this correct? */
278 qsize
= atomic_read(&__task_cred(p
)->user
->sigpending
);
280 qlim
= task_rlimit(p
, RLIMIT_SIGPENDING
);
281 unlock_task_sighand(p
, &flags
);
284 seq_printf(m
, "Threads:\t%d\n", num_threads
);
285 seq_printf(m
, "SigQ:\t%lu/%lu\n", qsize
, qlim
);
287 /* render them all */
288 render_sigset_t(m
, "SigPnd:\t", &pending
);
289 render_sigset_t(m
, "ShdPnd:\t", &shpending
);
290 render_sigset_t(m
, "SigBlk:\t", &blocked
);
291 render_sigset_t(m
, "SigIgn:\t", &ignored
);
292 render_sigset_t(m
, "SigCgt:\t", &caught
);
295 static void render_cap_t(struct seq_file
*m
, const char *header
,
301 CAP_FOR_EACH_U32(__capi
) {
302 seq_printf(m
, "%08x",
303 a
->cap
[CAP_LAST_U32
- __capi
]);
308 static inline void task_cap(struct seq_file
*m
, struct task_struct
*p
)
310 const struct cred
*cred
;
311 kernel_cap_t cap_inheritable
, cap_permitted
, cap_effective
,
312 cap_bset
, cap_ambient
;
315 cred
= __task_cred(p
);
316 cap_inheritable
= cred
->cap_inheritable
;
317 cap_permitted
= cred
->cap_permitted
;
318 cap_effective
= cred
->cap_effective
;
319 cap_bset
= cred
->cap_bset
;
320 cap_ambient
= cred
->cap_ambient
;
323 render_cap_t(m
, "CapInh:\t", &cap_inheritable
);
324 render_cap_t(m
, "CapPrm:\t", &cap_permitted
);
325 render_cap_t(m
, "CapEff:\t", &cap_effective
);
326 render_cap_t(m
, "CapBnd:\t", &cap_bset
);
327 render_cap_t(m
, "CapAmb:\t", &cap_ambient
);
330 static inline void task_seccomp(struct seq_file
*m
, struct task_struct
*p
)
332 #ifdef CONFIG_SECCOMP
333 seq_printf(m
, "Seccomp:\t%d\n", p
->seccomp
.mode
);
337 static inline void task_context_switch_counts(struct seq_file
*m
,
338 struct task_struct
*p
)
340 seq_printf(m
, "voluntary_ctxt_switches:\t%lu\n"
341 "nonvoluntary_ctxt_switches:\t%lu\n",
346 static void task_cpus_allowed(struct seq_file
*m
, struct task_struct
*task
)
348 seq_printf(m
, "Cpus_allowed:\t%*pb\n",
349 cpumask_pr_args(&task
->cpus_allowed
));
350 seq_printf(m
, "Cpus_allowed_list:\t%*pbl\n",
351 cpumask_pr_args(&task
->cpus_allowed
));
354 int proc_pid_status(struct seq_file
*m
, struct pid_namespace
*ns
,
355 struct pid
*pid
, struct task_struct
*task
)
357 struct mm_struct
*mm
= get_task_mm(task
);
360 task_state(m
, ns
, pid
, task
);
368 task_seccomp(m
, task
);
369 task_cpus_allowed(m
, task
);
370 cpuset_task_status_allowed(m
, task
);
371 task_context_switch_counts(m
, task
);
375 static int do_task_stat(struct seq_file
*m
, struct pid_namespace
*ns
,
376 struct pid
*pid
, struct task_struct
*task
, int whole
)
378 unsigned long vsize
, eip
, esp
, wchan
= 0;
380 int tty_pgrp
= -1, tty_nr
= 0;
381 sigset_t sigign
, sigcatch
;
383 pid_t ppid
= 0, pgid
= -1, sid
= -1;
386 struct mm_struct
*mm
;
387 unsigned long long start_time
;
388 unsigned long cmin_flt
= 0, cmaj_flt
= 0;
389 unsigned long min_flt
= 0, maj_flt
= 0;
390 cputime_t cutime
, cstime
, utime
, stime
;
391 cputime_t cgtime
, gtime
;
392 unsigned long rsslim
= 0;
393 char tcomm
[sizeof(task
->comm
)];
396 state
= *get_task_state(task
);
397 vsize
= eip
= esp
= 0;
398 permitted
= ptrace_may_access(task
, PTRACE_MODE_READ
| PTRACE_MODE_NOAUDIT
);
399 mm
= get_task_mm(task
);
401 vsize
= task_vsize(mm
);
403 eip
= KSTK_EIP(task
);
404 esp
= KSTK_ESP(task
);
408 get_task_comm(tcomm
, task
);
410 sigemptyset(&sigign
);
411 sigemptyset(&sigcatch
);
412 cutime
= cstime
= utime
= stime
= 0;
415 if (lock_task_sighand(task
, &flags
)) {
416 struct signal_struct
*sig
= task
->signal
;
419 struct pid
*pgrp
= tty_get_pgrp(sig
->tty
);
420 tty_pgrp
= pid_nr_ns(pgrp
, ns
);
422 tty_nr
= new_encode_dev(tty_devnum(sig
->tty
));
425 num_threads
= get_nr_threads(task
);
426 collect_sigign_sigcatch(task
, &sigign
, &sigcatch
);
428 cmin_flt
= sig
->cmin_flt
;
429 cmaj_flt
= sig
->cmaj_flt
;
430 cutime
= sig
->cutime
;
431 cstime
= sig
->cstime
;
432 cgtime
= sig
->cgtime
;
433 rsslim
= ACCESS_ONCE(sig
->rlim
[RLIMIT_RSS
].rlim_cur
);
435 /* add up live thread stats at the group level */
437 struct task_struct
*t
= task
;
439 min_flt
+= t
->min_flt
;
440 maj_flt
+= t
->maj_flt
;
441 gtime
+= task_gtime(t
);
442 } while_each_thread(task
, t
);
444 min_flt
+= sig
->min_flt
;
445 maj_flt
+= sig
->maj_flt
;
446 thread_group_cputime_adjusted(task
, &utime
, &stime
);
450 sid
= task_session_nr_ns(task
, ns
);
451 ppid
= task_tgid_nr_ns(task
->real_parent
, ns
);
452 pgid
= task_pgrp_nr_ns(task
, ns
);
454 unlock_task_sighand(task
, &flags
);
457 if (permitted
&& (!whole
|| num_threads
< 2))
458 wchan
= get_wchan(task
);
460 min_flt
= task
->min_flt
;
461 maj_flt
= task
->maj_flt
;
462 task_cputime_adjusted(task
, &utime
, &stime
);
463 gtime
= task_gtime(task
);
466 /* scale priority and nice values from timeslices to -20..20 */
467 /* to make it look like a "normal" Unix priority/nice value */
468 priority
= task_prio(task
);
469 nice
= task_nice(task
);
471 /* convert nsec -> ticks */
472 start_time
= nsec_to_clock_t(task
->real_start_time
);
474 seq_printf(m
, "%d (%s) %c", pid_nr_ns(pid
, ns
), tcomm
, state
);
475 seq_put_decimal_ll(m
, ' ', ppid
);
476 seq_put_decimal_ll(m
, ' ', pgid
);
477 seq_put_decimal_ll(m
, ' ', sid
);
478 seq_put_decimal_ll(m
, ' ', tty_nr
);
479 seq_put_decimal_ll(m
, ' ', tty_pgrp
);
480 seq_put_decimal_ull(m
, ' ', task
->flags
);
481 seq_put_decimal_ull(m
, ' ', min_flt
);
482 seq_put_decimal_ull(m
, ' ', cmin_flt
);
483 seq_put_decimal_ull(m
, ' ', maj_flt
);
484 seq_put_decimal_ull(m
, ' ', cmaj_flt
);
485 seq_put_decimal_ull(m
, ' ', cputime_to_clock_t(utime
));
486 seq_put_decimal_ull(m
, ' ', cputime_to_clock_t(stime
));
487 seq_put_decimal_ll(m
, ' ', cputime_to_clock_t(cutime
));
488 seq_put_decimal_ll(m
, ' ', cputime_to_clock_t(cstime
));
489 seq_put_decimal_ll(m
, ' ', priority
);
490 seq_put_decimal_ll(m
, ' ', nice
);
491 seq_put_decimal_ll(m
, ' ', num_threads
);
492 seq_put_decimal_ull(m
, ' ', 0);
493 seq_put_decimal_ull(m
, ' ', start_time
);
494 seq_put_decimal_ull(m
, ' ', vsize
);
495 seq_put_decimal_ull(m
, ' ', mm
? get_mm_rss(mm
) : 0);
496 seq_put_decimal_ull(m
, ' ', rsslim
);
497 seq_put_decimal_ull(m
, ' ', mm
? (permitted
? mm
->start_code
: 1) : 0);
498 seq_put_decimal_ull(m
, ' ', mm
? (permitted
? mm
->end_code
: 1) : 0);
499 seq_put_decimal_ull(m
, ' ', (permitted
&& mm
) ? mm
->start_stack
: 0);
500 seq_put_decimal_ull(m
, ' ', esp
);
501 seq_put_decimal_ull(m
, ' ', eip
);
502 /* The signal information here is obsolete.
503 * It must be decimal for Linux 2.0 compatibility.
504 * Use /proc/#/status for real-time signals.
506 seq_put_decimal_ull(m
, ' ', task
->pending
.signal
.sig
[0] & 0x7fffffffUL
);
507 seq_put_decimal_ull(m
, ' ', task
->blocked
.sig
[0] & 0x7fffffffUL
);
508 seq_put_decimal_ull(m
, ' ', sigign
.sig
[0] & 0x7fffffffUL
);
509 seq_put_decimal_ull(m
, ' ', sigcatch
.sig
[0] & 0x7fffffffUL
);
512 * We used to output the absolute kernel address, but that's an
513 * information leak - so instead we show a 0/1 flag here, to signal
514 * to user-space whether there's a wchan field in /proc/PID/wchan.
516 * This works with older implementations of procps as well.
523 seq_put_decimal_ull(m
, ' ', 0);
524 seq_put_decimal_ull(m
, ' ', 0);
525 seq_put_decimal_ll(m
, ' ', task
->exit_signal
);
526 seq_put_decimal_ll(m
, ' ', task_cpu(task
));
527 seq_put_decimal_ull(m
, ' ', task
->rt_priority
);
528 seq_put_decimal_ull(m
, ' ', task
->policy
);
529 seq_put_decimal_ull(m
, ' ', delayacct_blkio_ticks(task
));
530 seq_put_decimal_ull(m
, ' ', cputime_to_clock_t(gtime
));
531 seq_put_decimal_ll(m
, ' ', cputime_to_clock_t(cgtime
));
533 if (mm
&& permitted
) {
534 seq_put_decimal_ull(m
, ' ', mm
->start_data
);
535 seq_put_decimal_ull(m
, ' ', mm
->end_data
);
536 seq_put_decimal_ull(m
, ' ', mm
->start_brk
);
537 seq_put_decimal_ull(m
, ' ', mm
->arg_start
);
538 seq_put_decimal_ull(m
, ' ', mm
->arg_end
);
539 seq_put_decimal_ull(m
, ' ', mm
->env_start
);
540 seq_put_decimal_ull(m
, ' ', mm
->env_end
);
542 seq_printf(m
, " 0 0 0 0 0 0 0");
545 seq_put_decimal_ll(m
, ' ', task
->exit_code
);
547 seq_put_decimal_ll(m
, ' ', 0);
555 int proc_tid_stat(struct seq_file
*m
, struct pid_namespace
*ns
,
556 struct pid
*pid
, struct task_struct
*task
)
558 return do_task_stat(m
, ns
, pid
, task
, 0);
561 int proc_tgid_stat(struct seq_file
*m
, struct pid_namespace
*ns
,
562 struct pid
*pid
, struct task_struct
*task
)
564 return do_task_stat(m
, ns
, pid
, task
, 1);
567 int proc_pid_statm(struct seq_file
*m
, struct pid_namespace
*ns
,
568 struct pid
*pid
, struct task_struct
*task
)
570 unsigned long size
= 0, resident
= 0, shared
= 0, text
= 0, data
= 0;
571 struct mm_struct
*mm
= get_task_mm(task
);
574 size
= task_statm(mm
, &shared
, &text
, &data
, &resident
);
578 * For quick read, open code by putting numbers directly
580 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n",
581 * size, resident, shared, text, data);
583 seq_put_decimal_ull(m
, 0, size
);
584 seq_put_decimal_ull(m
, ' ', resident
);
585 seq_put_decimal_ull(m
, ' ', shared
);
586 seq_put_decimal_ull(m
, ' ', text
);
587 seq_put_decimal_ull(m
, ' ', 0);
588 seq_put_decimal_ull(m
, ' ', data
);
589 seq_put_decimal_ull(m
, ' ', 0);
595 #ifdef CONFIG_PROC_CHILDREN
597 get_children_pid(struct inode
*inode
, struct pid
*pid_prev
, loff_t pos
)
599 struct task_struct
*start
, *task
;
600 struct pid
*pid
= NULL
;
602 read_lock(&tasklist_lock
);
604 start
= pid_task(proc_pid(inode
), PIDTYPE_PID
);
609 * Lets try to continue searching first, this gives
610 * us significant speedup on children-rich processes.
613 task
= pid_task(pid_prev
, PIDTYPE_PID
);
614 if (task
&& task
->real_parent
== start
&&
615 !(list_empty(&task
->sibling
))) {
616 if (list_is_last(&task
->sibling
, &start
->children
))
618 task
= list_first_entry(&task
->sibling
,
619 struct task_struct
, sibling
);
620 pid
= get_pid(task_pid(task
));
628 * We might miss some children here if children
629 * are exited while we were not holding the lock,
630 * but it was never promised to be accurate that
633 * "Just suppose that the parent sleeps, but N children
634 * exit after we printed their tids. Now the slow paths
635 * skips N extra children, we miss N tasks." (c)
637 * So one need to stop or freeze the leader and all
638 * its children to get a precise result.
640 list_for_each_entry(task
, &start
->children
, sibling
) {
642 pid
= get_pid(task_pid(task
));
648 read_unlock(&tasklist_lock
);
652 static int children_seq_show(struct seq_file
*seq
, void *v
)
654 struct inode
*inode
= seq
->private;
657 pid
= pid_nr_ns(v
, inode
->i_sb
->s_fs_info
);
658 seq_printf(seq
, "%d ", pid
);
663 static void *children_seq_start(struct seq_file
*seq
, loff_t
*pos
)
665 return get_children_pid(seq
->private, NULL
, *pos
);
668 static void *children_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
672 pid
= get_children_pid(seq
->private, v
, *pos
+ 1);
679 static void children_seq_stop(struct seq_file
*seq
, void *v
)
684 static const struct seq_operations children_seq_ops
= {
685 .start
= children_seq_start
,
686 .next
= children_seq_next
,
687 .stop
= children_seq_stop
,
688 .show
= children_seq_show
,
691 static int children_seq_open(struct inode
*inode
, struct file
*file
)
696 ret
= seq_open(file
, &children_seq_ops
);
700 m
= file
->private_data
;
706 int children_seq_release(struct inode
*inode
, struct file
*file
)
708 seq_release(inode
, file
);
712 const struct file_operations proc_tid_children_operations
= {
713 .open
= children_seq_open
,
716 .release
= children_seq_release
,
718 #endif /* CONFIG_PROC_CHILDREN */